CN114387232A - Wafer center positioning, wafer gap positioning and wafer positioning calibration method - Google Patents

The invention provides a method for positioning a wafer center, positioning a wafer notch and positioning and calibrating a wafer, which comprises the following steps: the method comprises the steps of obtaining a wafer circumference image, carrying out binarization processing on the image by using a deep learning semantic segmentation model, screening contour lines to obtain a wafer edge, obtaining a wafer notch image, carrying out binarization processing on the image by using the deep learning semantic segmentation model, screening the contour lines to obtain a wafer notch edge curve, further obtaining a wafer notch top edge, and fitting the wafer center and the notch center to complete wafer positioning. The positioning method based on Deep learning (Deep learning) semantic segmentation has the advantages of high precision, no need of prior information, high automation, capability of effectively improving the positioning precision of the wafer and capability of accurately measuring the circle center and the notch of the wafer.

一种晶圆圆心定位、晶圆缺口定位及晶圆定位校准方法A wafer center positioning, wafer gap positioning and wafer positioning calibration method

技术领域technical field

本发明涉及半导体制造领域，更具体地，涉及一种晶圆圆心定位、晶圆缺口定位及晶圆定位校准方法。The invention relates to the field of semiconductor manufacturing, and more particularly, to a method for wafer center positioning, wafer gap positioning and wafer positioning calibration.

背景技术Background technique

晶圆定位是集成电路的制造和检测领域中重要的环节。由于集成电路的缺陷检测、尺寸测量等需要在特定位置进行测量，需要保证该位置出现在测量设备的视野(FOV，Field of View)中，因此需要对晶圆进行定位。具体来说，是通过一定的方法，测量得到晶圆的圆心和缺口的位置，然后计算晶圆相对机台的旋转角，即可得到晶圆与机台的坐标转换关系，即包含晶圆圆心定位和晶圆缺口定位两个步骤。Wafer positioning is an important link in the field of integrated circuit manufacturing and inspection. Since defect detection and size measurement of integrated circuits need to be measured at a specific position, it is necessary to ensure that the position appears in the field of view (FOV, Field of View) of the measurement equipment, so the wafer needs to be positioned. Specifically, through a certain method, the center of the wafer and the position of the gap are measured, and then the rotation angle of the wafer relative to the machine is calculated to obtain the coordinate conversion relationship between the wafer and the machine, that is, the center of the wafer is included. Positioning and wafer notch positioning are two steps.

目前，大多数的晶圆定位方法采用CCD等光学探测器作为图像采集设备，由于光学探测器的FOV较大，晶圆定位的精度不高。在后续使用较小的FOV检测缺陷时，定位的误差会导致缺陷偏离检测设备视野，最终导致缺陷漏检。对于刻有图形的晶圆，可以通过晶圆上的特殊图形进行更加准确的定位较准。但是，对于未刻图形的晶圆，无法获取额外的信息对定位进行校准。因此，需要提高晶圆圆心定位和缺口定位的精度。At present, most wafer positioning methods use optical detectors such as CCDs as image acquisition devices. Due to the large FOV of the optical detectors, the wafer positioning accuracy is not high. When a smaller FOV is used to detect defects later, the positioning error will cause the defects to deviate from the field of view of the inspection equipment, and eventually lead to missed defects. For wafers engraved with patterns, more accurate positioning can be performed through special patterns on the wafer. However, for unpatterned wafers, no additional information is available to calibrate the positioning. Therefore, it is necessary to improve the accuracy of wafer center positioning and notch positioning.

发明内容SUMMARY OF THE INVENTION

本发明针对现有技术中存在的技术问题，提供一种晶圆圆心定位、晶圆缺口定位及晶圆定位校准方法。Aiming at the technical problems existing in the prior art, the present invention provides a wafer center positioning, wafer gap positioning and wafer positioning calibration methods.

根据本发明的第一方面，提供了一种晶圆圆心定位方法，包括：According to a first aspect of the present invention, there is provided a method for locating the center of a wafer, comprising:

获取晶圆圆周图像；Obtain wafer circumference images;

基于深度学习语义分割模型对所述晶圆圆周图像进行二值化处理，提取所述晶圆圆周图像中的所有轮廓线；Binarization is performed on the wafer circumference image based on a deep learning semantic segmentation model, and all contour lines in the wafer circumference image are extracted;

从所有轮廓线中筛选出晶圆边缘轮廓线；Filter out wafer edge contours from all contours;

基于所述晶圆边缘轮廓线，拟合得到晶圆圆心坐标。Based on the contour line of the wafer edge, the coordinates of the center of the wafer are obtained by fitting.

在上述技术方案的基础上，本发明还可以作出如下改进。On the basis of the above technical solutions, the present invention can also make the following improvements.

可选的，所述从所有轮廓线中筛选出晶圆边缘轮廓线，包括：Optionally, the wafer edge contour lines are screened out from all contour lines, including:

获取所有轮廓线的长度以及所述晶圆圆周图像的宽度或高度；Obtain the lengths of all contour lines and the width or height of the wafer circumference image;

筛选满足如下条件的轮廓线为所述晶圆边缘轮廓线；Screening contour lines that meet the following conditions are the wafer edge contour lines;

轮廓线的长度与所述晶圆圆周图像的宽度或高度的比值大于1且小于1.1。The ratio of the length of the contour line to the width or height of the wafer circumference image is greater than 1 and less than 1.1.

可选的，所述提取每一条轮廓线的轮廓特征，包括：Optionally, the extracting contour features of each contour line includes:

基于任一条轮廓线上所有边缘点的横坐标、纵坐标、以及所述晶圆圆周图像的宽度和高度，计算所述任一条轮廓线上边缘点的离散程度；Calculate the degree of dispersion of edge points on any contour line based on the abscissa and ordinate of all edge points on any contour line, and the width and height of the wafer circumference image;

根据所述任一条轮廓线上所有边缘点的横纵坐标以及轮廓线的二阶梯度，计算所述任一条轮廓线的弯曲程度；According to the horizontal and vertical coordinates of all edge points on the any contour line and the second-order gradient of the contour line, calculate the degree of curvature of the any one contour line;

根据所述任一条轮廓线上所有边缘点的横纵坐标，拟合得到直线方程，基于所述直线方程，计算直线拟合的残差平方和作为所述任一条轮廓线近似直线程度；According to the horizontal and vertical coordinates of all edge points on the any contour line, a straight line equation is obtained by fitting, and based on the straight line equation, the residual sum of squares of the straight line fitting is calculated as the approximate straight line degree of any one contour line;

相应的，所述基于每一条轮廓线的轮廓特征，从所有轮廓线中筛选出晶圆边缘轮廓线，包括：Correspondingly, based on the contour features of each contour line, the wafer edge contour lines are screened from all contour lines, including:

以所述任一条轮廓线上边缘点的离散程度、所述任一条轮廓线的弯曲程度以及所述任一条轮廓线近似直线程度中的一个或两个以上的组合的和作为所述任一条轮廓线的评分；Taking the discrete degree of edge points on any one of the contour lines, the degree of bending of the any one of the contour lines, and the degree of the approximate straight line of the any one of the contour lines. line rating;

将评分最低的轮廓线作为晶圆边缘轮廓线。Take the contour with the lowest score as the wafer edge contour.

可选的，所述基于所述晶圆边缘轮廓线，拟合得到晶圆圆心坐标，还包括：Optionally, according to the outline of the wafer edge, the coordinates of the center of the wafer are obtained by fitting, further comprising:

将所述晶圆边缘轮廓线切割为多段子边缘线；cutting the wafer edge contour line into multiple sub-edge lines;

在所述多段子边缘线上选取多个边缘点；Select a plurality of edge points on the multiple sub-edge lines;

基于选取的多个边缘点，拟合得到晶圆圆心坐标。Based on the selected edge points, the coordinates of the center of the wafer are obtained by fitting.

根据本发明的第二方面，提供一种晶圆缺口定位方法，包括：According to a second aspect of the present invention, there is provided a wafer gap positioning method, comprising:

获取晶圆缺口图像；Obtain wafer notch images;

基于深度学习语义分割模型对所述晶圆缺口图像进行二值化，提取所述晶圆缺口图像中的所有轮廓线；Binarize the wafer gap image based on a deep learning semantic segmentation model, and extract all contour lines in the wafer gap image;

从所有轮廓线中筛选出晶圆缺口边缘曲线；Screen out the wafer notch edge curve from all contour lines;

从所述晶圆缺口边缘曲线确定晶圆缺口的顶部边缘曲线；determining the top edge curve of the wafer notch from the wafer notch edge curve;

基于所述晶圆缺口的顶部边缘曲线，拟合得到晶圆缺口圆心坐标。Based on the top edge curve of the wafer gap, the coordinates of the center of the wafer gap are obtained by fitting.

可选的，所述从所有轮廓线中筛选出晶圆缺口边缘曲线，包括：Optionally, the wafer notch edge curve is screened from all contour lines, including:

提取每一条轮廓线的轮廓特征，所述轮廓特征包括轮廓线的弯曲程度和近似二次曲线程度中任意一个或两个的组合；Extracting the contour feature of each contour line, the contour feature includes any one or a combination of the curve degree of the contour line and the approximate quadratic curve degree;

基于每一条轮廓线的轮廓特征，从所有轮廓线中筛选出晶圆缺口边缘曲线。Based on the contour features of each contour line, the wafer notch edge curve is screened from all contour lines.

可选的，所述提取每一条轮廓线的轮廓特征，包括：Optionally, the extracting contour features of each contour line includes:

根据任一条轮廓线上所有边缘点的横纵坐标及轮廓线的二阶梯度，计算所述任一条轮廓线的弯曲程度；According to the abscissa and vertical coordinates of all edge points on any contour line and the second-order gradient of the contour line, calculate the bending degree of any one contour line;

根据所述任一条轮廓线上所有边缘点的横纵坐标，拟合得到二次曲线方程，基于所述二次曲线方程，计算二次曲线拟合的残差平方和作为所述任一条轮廓线的近似二次曲线程度；According to the horizontal and vertical coordinates of all edge points on any contour line, a quadratic curve equation is obtained by fitting, and based on the quadratic curve equation, the residual sum of squares of the quadratic curve fitting is calculated as the any contour line The approximate degree of quadratic curve;

相应的，基于每一条轮廓线的轮廓特征，从所有轮廓线中筛选出晶圆缺口边缘曲线，包括：Correspondingly, based on the contour features of each contour line, the wafer notch edge curve is screened from all contour lines, including:

所述任一条轮廓线的弯曲程度以及所述任一条轮廓线的近似二次曲线程度中一个或两个组合的和作为所述任一条轮廓线的评分；The sum of one or two combinations of the degree of curvature of the any one of the contour lines and the degree of the approximate quadratic curve of the any one of the contour lines is used as the score of the any one of the contour lines;

将评分最低的轮廓线作为晶圆缺口边缘轮廓线。Take the contour with the lowest score as the wafer notch edge contour.

可选的，所述从所述晶圆缺口边缘曲线确定晶圆缺口的顶部边缘曲线，包括：Optionally, the determining the top edge curve of the wafer gap from the wafer gap edge curve includes:

获取所述晶圆缺口边缘曲线的所有边缘点的中心点，以所述中心点为基准，截取所述晶圆缺口边缘曲线，将所述晶圆缺口边缘曲线两侧边缘部分剔除，得到所述晶圆缺口的顶部边缘曲线。Obtain the center points of all edge points of the edge curve of the wafer gap, take the center point as a benchmark, intercept the edge curve of the wafer gap, and remove the edge portions on both sides of the edge curve of the wafer gap to obtain the The top edge curve of the wafer notch.

可选的，基于所述晶圆缺口的顶部边缘曲线，拟合得到晶圆缺口圆心坐标，包括：Optionally, based on the top edge curve of the wafer gap, the coordinates of the center of the wafer gap are obtained by fitting, including:

获取所述晶圆缺口的顶部边缘曲线上的所有顶部边缘点，构建多个顶部边缘点集合，且每个顶部边缘点集合包含的顶部边缘点不重合；Obtaining all top edge points on the top edge curve of the wafer gap, constructing multiple top edge point sets, and the top edge points included in each top edge point set do not overlap;

对每个顶部边缘点集合中的边缘点，拟合得到对应晶圆缺口圆心坐标；For each edge point in the top edge point set, fit the coordinates of the center of the corresponding wafer gap;

将所有晶圆缺口圆心坐标求取平均值，得到最终的晶圆缺口圆心坐标。The coordinates of the center of all wafer gaps are averaged to obtain the final coordinates of the center of the wafer gap.

根据本发明的第三方面，提供了一种晶圆定位校准方法，包括：According to a third aspect of the present invention, a wafer positioning calibration method is provided, comprising:

基于晶圆圆心定位方法，获取所述晶圆圆心坐标；Based on the wafer center positioning method, obtain the coordinates of the wafer center;

基于晶圆缺口定位方法，获取所述晶圆缺口的圆心坐标；Based on the wafer gap positioning method, obtain the coordinates of the center of the wafer gap;

基于所述晶圆圆心坐标和所述晶圆缺口的圆心坐标，获取晶圆旋转角，完成晶圆定位校准。Based on the coordinates of the center of the wafer and the coordinates of the center of the wafer gap, the rotation angle of the wafer is obtained, and the wafer positioning calibration is completed.

本发明提供的一种晶圆圆心定位、晶圆缺口定位及晶圆定位校准方法，获取晶圆圆周图像，使用深度学习语义分割模型对图像做二值化处理，筛选轮廓线得到晶圆边缘，获取晶圆缺口图像，使用深度学习语义分割模型对图像做二值化处理，筛选轮廓线得到晶圆缺口边缘曲线，确定晶圆缺口的顶部边缘曲线，拟合晶圆圆心和缺口圆心完成晶圆定位。本发明基于深度学习(Deep learning)语义分割的定位方法，精度高，无需先验信息，自动化高，能有效的提高晶圆定位精度，可以精确的测量晶圆圆心和缺口。The invention provides a wafer center positioning, wafer gap positioning and wafer positioning calibration method, obtaining a wafer circumference image, using a deep learning semantic segmentation model to perform binarization processing on the image, filtering contour lines to obtain the wafer edge, Obtain the wafer gap image, use the deep learning semantic segmentation model to binarize the image, filter the contour lines to obtain the wafer gap edge curve, determine the top edge curve of the wafer gap, fit the wafer center and the gap center to complete the wafer position. The present invention is based on the deep learning (Deep learning) semantic segmentation positioning method, has high precision, does not require prior information, is highly automated, can effectively improve the wafer positioning precision, and can accurately measure the wafer center and gap.

附图说明Description of drawings

图1为本发明提供的一种晶圆圆心定位方法流程图；1 is a flowchart of a method for locating the center of a wafer provided by the present invention;

图2为晶圆圆心拟合示意图；Fig. 2 is a schematic diagram of wafer center fitting;

图3为晶圆缺口定位方法流程图；3 is a flowchart of a wafer gap positioning method;

图4为晶圆缺口粗定位中获取晶圆缺口顶部示意图；4 is a schematic diagram of obtaining the top of the wafer gap in the rough positioning of the wafer gap;

图5为晶圆缺口圆形拟合示意图；FIG. 5 is a schematic diagram of circular fitting of a wafer gap;

图6为本发明提供的一种晶圆圆心定位系统的结构示意图；6 is a schematic structural diagram of a wafer center positioning system provided by the present invention;

图7为本发明提供的一种晶圆缺口定位系统的结构示意图；7 is a schematic structural diagram of a wafer gap positioning system provided by the present invention;

图8为本发明提供的一种可能的电子设备的硬件结构示意图；8 is a schematic diagram of the hardware structure of a possible electronic device provided by the present invention;

图9为本发明提供的一种可能的计算机可读存储介质的硬件结构示意图。FIG. 9 is a schematic diagram of the hardware structure of a possible computer-readable storage medium provided by the present invention.

具体实施方式Detailed ways

下面结合附图和实施例，对本发明的具体实施方式作进一步详细描述。以下实施例用于说明本发明，但不用来限制本发明的范围。The specific embodiments of the present invention will be described in further detail below with reference to the accompanying drawings and embodiments. The following examples are intended to illustrate the present invention, but not to limit the scope of the present invention.

在晶圆圆心定位和晶圆缺口定位的实现过程中，采用CCD等光学探测器通过测量晶圆圆心和缺口进行晶圆定位得到的定位精度不高。对于无图形晶圆，由于晶圆中没有特殊图形，无法像有图形晶圆一样对定位再次校准得到高精度的定位。本发明提出了一种基于深度学习语义分割模型的晶圆定位方法，提高了晶圆定位精度。通常采用高倍率、高分辨率成像设备对晶圆定位，例如带电粒子束扫描成像设备(例如SEM、FIB-SEM等)、AFM成像设备等，为便于说明，本发明实施例以扫描电子显微镜获取SEM图像为例说明，可以理解的是，SEM图像并不用来限制本发明的范围。In the realization process of wafer center positioning and wafer gap positioning, the positioning accuracy obtained by using an optical detector such as a CCD to locate the wafer by measuring the wafer center and gap is not high. For a non-patterned wafer, since there is no special pattern in the wafer, the positioning cannot be recalibrated to obtain high-precision positioning like a patterned wafer. The invention proposes a wafer positioning method based on a deep learning semantic segmentation model, which improves the wafer positioning accuracy. Usually, high-magnification, high-resolution imaging equipment is used to position the wafer, such as charged particle beam scanning imaging equipment (such as SEM, FIB-SEM, etc.), AFM imaging equipment, etc., for the convenience of description, the embodiment of the present invention is obtained by scanning electron microscope The SEM image is used as an example to illustrate, and it is understood that the SEM image is not used to limit the scope of the present invention.

实施例一Example 1

一种晶圆圆心定位方法，参见图1，该晶圆圆心定位方法主要包括以下步骤：A method for locating the center of a wafer, see FIG. 1 , the method for locating the center of a wafer mainly includes the following steps:

S1，获取晶圆圆周图像。S1, obtain a wafer circumference image.

本发明实施例采用扫描电子束显微镜SEM对晶圆进行扫描，得到晶圆圆周SEM图像，后续处理均是基于晶圆圆周SEM图像进行的。In the embodiment of the present invention, a scanning electron beam microscope SEM is used to scan the wafer to obtain a SEM image of the circumference of the wafer, and subsequent processing is performed based on the SEM image of the circumference of the wafer.

S2，基于深度学习语义分割模型对所述晶圆圆周图像进行二值化处理，提取所述晶圆圆周图像中的所有轮廓线。S2, perform binarization processing on the wafer circumference image based on a deep learning semantic segmentation model, and extract all contour lines in the wafer circumference image.

可以理解的是，当扫描得到晶圆圆周SEM图像，从中提取轮廓线。具体的，对晶圆圆周SEM图像进行二值化处理，基于二值化图像提取其中的所有轮廓线。可以理解的是，可以对晶圆SEM图像进行去噪处理，将图像大小缩放到深度神经网络(DNN)模型的输入大小，例如512*512，基于深度学习语义分割模型对晶圆SEM图像中进行二值化处理，基于二值化的晶圆圆周SEM图像，可以获取晶圆边缘轮廓线。It can be understood that when scanning a SEM image of the wafer circumference, contour lines are extracted from it. Specifically, a binarization process is performed on the SEM image of the wafer circumference, and all contour lines in the SEM image are extracted based on the binarized image. It can be understood that the wafer SEM image can be denoised, the image size can be scaled to the input size of the deep neural network (DNN) model, such as 512*512, and the wafer SEM image can be processed based on the deep learning semantic segmentation model. The binarization process, based on the binarized SEM image of the wafer circumference, can obtain the contour of the wafer edge.

所述晶圆圆周图像具有图像噪声大、模糊、明暗对比度变化大等特点。晶圆圆周图像中边缘部分可能存在缺陷，也可能出现与晶圆纹理相似的机台部分。使用传统计算机视觉方法，基于像素或者提取纹理等低层特征进行图像分割难以准确的获得晶圆边缘。而本发明使用深度学习语义分割模型对晶圆圆周图像进行二值化分割，由于深度语义分割模型提取了不同尺度、不同语义层次的图像特征，分割的边缘更加准确，平滑，且模型的鲁棒性高，泛化能力强，从而提高了晶圆圆心定位的准确性。The wafer circumference image has the characteristics of large image noise, blur, and large change in contrast between light and dark. There may be defects in the edge portion of the wafer circumference image, and there may be a part of the tool that is similar to the wafer texture. Using traditional computer vision methods, it is difficult to accurately obtain wafer edges for image segmentation based on low-level features such as pixels or extracted textures. The present invention uses the deep learning semantic segmentation model to perform binarization segmentation on the wafer circumference image. Since the deep semantic segmentation model extracts image features of different scales and different semantic levels, the edge of the segmentation is more accurate and smooth, and the model is robust. It has high performance and strong generalization ability, thus improving the accuracy of wafer center positioning.

在一些实施例中，获取晶圆边缘轮廓线包括提取其中的所有轮廓线，从所有轮廓线中筛选出晶圆边缘轮廓线。In some embodiments, obtaining the wafer edge contours includes extracting all the contours therein, and filtering out the wafer edge contours from all the contours.

作为实施例，获取各个轮廓线的轮廓特征，所述轮廓特征包括轮廓线的长度、轮廓线上边缘点的离散程度、轮廓线的弯曲程度、轮廓线近似直线程度中的任意一个或两个以上的组合，基于轮廓特征从所有轮廓线中筛选出晶圆边缘轮廓线。As an embodiment, the contour features of each contour line are acquired, where the contour features include any one or two or more of the length of the contour line, the degree of dispersion of edge points on the contour line, the degree of curvature of the contour line, and the approximate straight line degree of the contour line A combination of , filter out wafer edge contours from all contours based on contour features.

作为实施例，获取所有轮廓线的长度以及晶圆圆周图像的宽度或高度；轮廓线的长度与晶圆圆周图像的宽度或高度的比值大于1且小于1.1，则该轮廓线被筛选为晶圆边缘轮廓线。As an example, the lengths of all contour lines and the width or height of the wafer circumference image are obtained; the ratio of the length of the contour lines to the width or height of the wafer circumference image is greater than 1 and less than 1.1, then the contour lines are screened as wafers Edge contour lines.

即，轮廓线长度小于晶圆圆周图像的宽度或高度的被筛除，轮廓线长度远长于晶圆圆周图像的宽度或高度的被筛除，轮廓线上边缘点越离散被筛除，轮廓线越弯曲被筛除，轮廓线不近似直线被筛除。That is, those whose contour line length is less than the width or height of the wafer circumference image are screened out, those whose contour line length is much longer than the width or height of the wafer circumference image are screened out, and the more discrete the edge points on the contour line are, the more discrete the contour line The more curved lines are screened out, the contour lines that are not nearly straight are screened out.

可以理解的是，晶圆圆周SEM图像中包含晶圆边缘和机台部分，晶圆边缘中可能存在长条形的缺陷，机台部分也可能存在明显的轮廓纹理。这些缺陷、机台中的纹理会出现在二值化图像中，影响晶圆边缘轮廓的识别。考虑到晶圆边缘为平滑的圆弧，而缺陷或者机台纹理都是不规则非平滑的含有凹凸的曲线，本发明提出了一种轮廓线的评分方法来筛选晶圆边缘。晶圆边缘轮廓线近似直线且其长度与晶圆圆周图像宽度或高度相近，设计的评分方法从边缘点的离散程度、轮廓线的弯曲程度和轮廓线近似直线程度三个方面中的至少一个或两个以上组合，评价筛选晶圆边缘轮廓线。这种评分筛选轮廓线得到晶圆边缘轮廓线的方法能够提高模型对晶圆、机台、不同成像质量(比如，对于低信噪比的图像处理保持高精度)的图像的泛化能力，提高提取晶圆边缘轮廓线的能力、精度。It can be understood that the SEM image of the wafer circumference includes the wafer edge and the machine part. There may be long-shaped defects in the wafer edge, and there may also be obvious contour textures in the machine part. These defects and textures in the machine will appear in the binarized image and affect the recognition of the wafer edge contour. Considering that the wafer edge is a smooth arc, and the defects or machine texture are irregular and non-smooth curves containing concave and convex, the present invention proposes a contour scoring method to screen the wafer edge. The contour of the wafer edge is approximately straight and its length is similar to the width or height of the wafer circumference image. The scoring method is designed from at least one of three aspects: the degree of dispersion of edge points, the degree of curvature of the contour, and the degree of approximate straightness of the contour. Two or more combinations are evaluated for screening wafer edge profiles. This method of scoring and screening contours to obtain wafer edge contours can improve the generalization ability of the model to images of wafers, machines, and images of different imaging qualities (for example, to maintain high precision for image processing with low signal-to-noise ratio), improve the Ability and precision to extract wafer edge contours.

作为实施例，所述基于轮廓线上边缘点的离散程度、轮廓线的弯曲程度和轮廓线近似直线程度中的任意一个或两个以上的组合，从所有轮廓线中筛选出晶圆边缘轮廓线，包括：对于任一条轮廓线，基于所述任一条轮廓线上所有边缘点的横坐标中的最大值和最小值、纵坐标中的最大值和最小值、以及所述晶圆圆周图像的宽度和高度，计算所述任一条轮廓线上边缘点的离散程度Score_d；根据所述任一条轮廓线上所有边缘点的横纵坐标以及轮廓线的二阶梯度计算所述任一条轮廓线的弯曲程度Score_c；根据所述任一条轮廓线上所有边缘点的横纵坐标，拟合得到直线方程，基于所述直线方程，计算直线拟合的残差平方和SSE_line作为所述任一条轮廓线近似直线程度；相应的，所述基于每一条轮廓线的轮廓特征，从所有轮廓线中筛选出晶圆边缘轮廓线，包括：以所述任一条轮廓线上边缘点的离散程度Score_d、所述任一条轮廓线的弯曲程度Score_c以及所述任一条轮廓线近似直线程度SSE_line中的一个或两个以上的组合的和作为所述任一条轮廓线的评分。As an embodiment, based on any one or a combination of two or more of the discrete degree of edge points on the contour line, the degree of curvature of the contour line, and the approximate straightness of the contour line, the wafer edge contour line is screened from all the contour lines , including: for any contour line, based on the maximum and minimum values in the abscissa, the maximum value and the minimum value in the ordinate, and the width of the wafer circumference image based on the abscissa of all edge points on the any contour line and height, calculate the discrete degree Score _d of the edge points on the any contour line; calculate the bending of the any contour line according to the horizontal and vertical coordinates of all edge points on the any contour line and the second-order gradient of the contour line Degree Score _c ; According to the abscissa and vertical coordinates of all edge points on the described any contour line, a straight line equation is obtained by fitting, and based on the straight line equation, the residual square sum SSE _line of the straight line fitting is calculated as the any one contour line Approximate straightness; Correspondingly, based on the contour feature of each contour, the wafer edge contour is screened from all contours, including: with the discrete degree Score _d of edge points on any contour line, all The combined sum of one or more of the degree of curvature of any one of the contour lines, Score _c , and the degree of approximate straightness of any one of the contour lines, SSE _line , is used as the score of any one of the contour lines.

可选的，基于所述任一条轮廓线上边缘点的离散程度、轮廓线的弯曲程度和近似直线程度中的任意一个，以轮廓线上边缘点的离散程度为例说明，所述任一条轮廓线的评分Score_edge＝Score_d，Score_edge为所述任一条轮廓线的评分；将评分最低的轮廓线作为晶圆边缘轮廓线。Optionally, based on any one of the degree of dispersion of edge points on the contour line, the degree of curvature of the contour line, and the degree of approximate straightness, and the degree of dispersion of edge points on the contour line is used as an example to illustrate, the degree of dispersion of the edge points on the contour line The score of the line is Score _edge =Score _d , Score _edge is the score of any one of the contour lines; the contour line with the lowest score is taken as the wafer edge contour line.

可选的，基于所述任一条轮廓线上边缘点的离散程度、轮廓线的弯曲程度和近似直线程度中的任意两个以上的组合，计算所述任一条轮廓线的评分，以三个组合为例：Score_edge＝λ_dScore_d+λ_cScore_c+λ_lineSSE_line；其中，λ_d、λ_c、λ_line分别为Score_d、Score_c和SSE_line的权重，本发明实施例中，λ_d、λ_c、λ_line均取1。Score_edge为所述任一条轮廓线的评分；将评分最低的轮廓线作为晶圆边缘轮廓线。Optionally, based on the combination of any two or more of the discrete degree of edge points on the contour line, the degree of curvature of the contour line, and the approximate straight line degree, the score of any contour line is calculated, and three combinations are used. For example: Score _edge =λ _d Score _d +λ _c Score _c +λ _line SSE _line ; wherein, λ _d , λ _c , and λ _line are the weights of Score _d , Score _c and SSE _line , respectively. In the embodiment of the present invention, λ _d , λ _c , and λ _line are all set to 1. Score _edge is the score of any one of the contour lines; the contour line with the lowest score is used as the wafer edge contour line.

具体的，对于步骤S2从晶圆圆周图像中提取的所有轮廓线，通过评分筛选出晶圆边缘轮廓线，主要是基于轮廓线上边缘点的离散程度、轮廓线的弯曲程度和轮廓线近似直线程度等三个方面对每一条轮廓线进行衡量。Specifically, for all the contour lines extracted from the wafer circumference image in step S2, the wafer edge contour lines are screened out by scoring, which is mainly based on the discrete degree of edge points on the contour line, the degree of curvature of the contour line and the approximate straight line of the contour line. Each contour line is measured in three aspects such as degree.

其中，每一条轮廓线由点列(x_i,y_i),i＝1,…,N成构成。X、Y分别表示点列所有的横坐标集合、点列所有纵坐标集合，X_i、Y_i为轮廓线第i点的x、y坐标。Among them, each contour line is composed of a point sequence (x _i , y _i ), i=1, . . . , N. X and Y respectively represent the set of all abscissas of the point column and the set of all the ordinates of the point column, and X _i and Y _i are the x and y coordinates of the ith point of the contour line.

作为实施例，轮廓线上边缘点的离散程度Score_d由公式(1)-(3)得到。As an example, the discrete degree Score _d of edge points on the contour line is obtained by formulas (1)-(3).

Gap_x＝MAX(X)-MIN(X) (1)；Gap _x =MAX(X)-MIN(X) (1);

其中，MAX(X)、MIN(X)为X中最大值和最小值。Among them, MAX(X) and MIN(X) are the maximum and minimum values in X.

Gap_y＝MAX(Y)-MIN(Y) (2)；Gap _y =MAX(Y)-MIN(Y) (2);

其中，MAX(Y)、MIN(Y)为Y中最大值和最小值。Among them, MAX(Y) and MIN(Y) are the maximum and minimum values in Y.

其中，STD(X)为X的标准差，STD(Y)为Y的标准差，W和H为图像的宽和高，Score_d衡量轮廓线点列在x，y方向的离散程度。Among them, STD(X) is the standard deviation of X, STD(Y) is the standard deviation of Y, W and H are the width and height of the image, and Score _d measures the discrete degree of the outline points in the x and y directions.

作为实施例，轮廓线的弯曲程度Score_c由公式(4)-(10)得到。As an example, the degree of curvature Score _c of the contour line is obtained by formulas (4)-(10).

Score_t＝-N/(T_x+T_y) (7)；Score _t = -N/(T _x +T _y ) (7);

其中，T_x，T_y从轮廓线上出现拐点的次数来衡量轮廓线在x、y方向上的弯曲程度。Among them, T _x and T _y measure the bending degree of the contour line in the x and y directions by the number of inflection points appearing from the contour line.

g_i＝(Y_i-Y_i-1)/(X_i-X_i-1) (8)；g _i =(Y _i -Y _i-1 )/(X _i -X _i-1 ) (8);

Score_c＝λ_tScore_t+λ_gScore_g (10)；Score _c =λ _t Score _t +λ _g Score _g (10);

其中，g_i为轮廓线的二阶梯度，Score_g从二阶梯度的角度来衡量轮廓线的弯曲程度。λ_t、λ_g分别为Score_t和Score_g的权重，λ_t+λ_g＝1，优选λ_t、λ_g均取0.5；Score_t表示以轮廓线上所有边缘点的横纵坐标取值变化衡量轮廓线的弯曲程度，Score_g表示以轮廓线二阶梯度的角度衡量所述任一条轮廓线的弯曲程度。Among them, _gi is the second-order gradient of the contour line, and Score _g measures the curvature of the contour line from the angle of the second-order gradient. λ _t and λ _g are the weights of Score _t and Score _g respectively, λ _t +λ _g =1, preferably λ _t and λ _g are both 0.5; Score _t represents the change in the abscissa and ordinate values of all edge points on the contour line The degree of curvature of the contour line is measured, and Score _g represents the degree of curvature of any contour line measured by the angle of the second-order gradient of the contour line.

作为实施例，直线拟合的残差平方和由公式(11)-(12)得到。As an example, the residual sum of squares of the line fit is obtained by equations (11)-(12).

其中，

分别为X、Y的算数平均值。

为使用最小二乘法拟合直线得到的直线方程。in,

are the arithmetic mean of X and Y, respectively.

Equation for a straight line obtained by fitting a straight line using the least squares method.

其中，SSE_line为最小二乘拟合的残差平方和，作为所述任一条轮廓线近似直线程度。Wherein, SSE _line is the residual sum of squares of the least squares fitting, which is used as the approximate straight line degree of any one of the contour lines.

以Score_d、Score_c和SSE_line三个中的任意一个或两个以上的组合，得到晶圆圆周轮廓线的评分Score_edge。示例性的，以三个组合为例，则Score_edge＝λ_dScore_d+λ_cScore_c+λ_lineSSE_line，其中，Use any one of Score _d , Score _c and SSE _line or a combination of two or more to obtain the score Score _edge of the contour of the wafer circumference. Exemplarily, taking three combinations as an example, Score _edge =λ _d Score _d +λ _c Score _c +λ _line SSE _line , where,

Score_c＝λ_tScore_t+λ_gScore_g，那么晶圆圆周轮廓线的评分Score_edge的表达式为：Score _c =λ _t Score _t +λ _g Score _g , then the expression of the score Score _edge of the contour of the wafer circumference is:

Score_edge＝λ_dScore_d+λ_tScore_t+λ_gScore_g+λ_lineSSE_line (13)；Score _edge =λ _d Score _d +λ _t Score _t +λ _g Score _g +λ _line SSE _line (13);

其中，λ_d、λ_t、λ_g、λ_line分别为Score_d、Score_t、Score_g和SSE_line的权重，本发明实施例中，λ_d、λ_c、λ_line取1，λ_t+λ_g＝1，优选的，λ_t、λ_g均取0.5。Score_edge为晶圆圆周轮廓线的评分公式。Among them, λ _d , λ _t , λ _g , and λ _line are the weights of Score _d , Score _t , Score _g and SSE _line respectively. In the embodiment of the present invention, λ _d , λ _c , and λ _line are taken as 1, and λ _t +λ _g =1, preferably, λ _t and λ _g are both set to be 0.5. Score _edge is the scoring formula for the contour of the wafer circumference.

对于从晶圆圆周SEM图像中提取的所有轮廓线，对每一条轮廓线可选择上述任意一方法进行评分，得到对应分值，将分值最小的轮廓线作为晶圆边缘轮廓线。上述方法能够准确提取晶圆边缘轮廓线，提高了晶圆边缘轮廓线识别精度。For all the contour lines extracted from the SEM image of the wafer circumference, any one of the above methods can be selected for scoring each contour line to obtain the corresponding score, and the contour line with the smallest score is used as the wafer edge contour line. The above method can accurately extract the contour line of the wafer edge and improve the recognition accuracy of the contour line of the wafer edge.

S3，基于所述晶圆边缘轮廓线，拟合得到晶圆圆心。S3, based on the contour line of the wafer edge, obtain the center of the wafer circle by fitting.

可选的，所述基于所述晶圆边缘轮廓线，拟合得到晶圆圆心坐标，包括：从所述晶圆边缘轮廓线获取至少三个边缘点坐标，根据所述至少三个边缘点坐标确定所述晶圆圆心坐标。Optionally, obtaining the coordinates of the center of the wafer by fitting based on the wafer edge contour line includes: acquiring at least three edge point coordinates from the wafer edge contour line, and obtaining at least three edge point coordinates according to the at least three edge point coordinates. Determine the coordinates of the center of the wafer.

作为实施例，所述基于所述晶圆边缘轮廓线，拟合得到晶圆圆心坐标，包括：将所述晶圆边缘轮廓线切割为多段子边缘线，基于与评价每一条轮廓线相同的方法获取每段子边缘线的评分Score_edge，从中确定多条最优子边缘线，示例性的，以晶圆边缘轮廓切割为10段子边缘线为例，选取其中5段评分Score_edge最低的为最优子边缘线；在所述多条最优子边缘线上选取边缘点，示例性的，比如，从选取的5段最优子边缘线的每一段子边缘线上选取2个边缘点，总共选取10个边缘点；基于选取的边缘点，拟合晶圆的圆形形状，得到晶圆圆心坐标。As an example, obtaining the coordinates of the center of the wafer by fitting based on the wafer edge contour line includes: cutting the wafer edge contour line into multiple sub-edge lines, based on the same method as evaluating each contour line Obtain the score Score _edge of each sub-edge line, and determine a plurality of optimal sub-edge lines from it. Exemplarily, take the wafer edge contour cut into 10 sub-edge lines as an example, and select 5 segments with the lowest Score _edge as the optimal sub-edge line; select edge points on the plurality of optimal sub-edge lines, for example, select 2 edge points from each sub-edge line of the selected 5 optimal sub-edge lines, and select a total of 2 edge points 10 edge points; based on the selected edge points, fit the circular shape of the wafer to obtain the coordinates of the center of the wafer.

可选的，获取多个晶圆圆周图像，优选获取至少三个晶圆圆周图像，采用上述方法对其进行处理，分别提取对应的边缘点，从而拟合晶圆的圆形形状，得到晶圆圆心坐标。以获取三个晶圆圆周图像为例，所述三个晶圆圆周图像可以是均匀分布在晶圆圆周上，即三个晶圆图像对应的晶圆圆周位置与晶圆中心所成圆心角均大致为120度，可选的，三个晶圆图像对应的晶圆圆周位置与晶圆中心所成圆心角最小为90度。Optionally, obtain a plurality of wafer circumference images, preferably at least three wafer circumference images, process them by the above method, and extract corresponding edge points respectively, so as to fit the circular shape of the wafer, and obtain the wafer. The coordinates of the center of the circle. Taking the acquisition of three wafer circumference images as an example, the three wafer circumference images may be evenly distributed on the wafer circumference, that is, the position of the wafer circumference corresponding to the three wafer images and the central angle formed by the center of the wafer are equal to each other. It is approximately 120 degrees. Optionally, the central angle formed between the circumferential positions of the wafers corresponding to the three wafer images and the center of the wafer is at least 90 degrees.

可以理解的是，对于S2中确定的晶圆边缘轮廓线，本步骤将晶圆边缘轮廓线切割为多段子边缘线，比如，将晶圆边缘轮廓线等分为多段子边缘线，并对每一段子边缘线进行评分，选取最优的多段子边缘线，即评分Score_edge最低的多段子边缘线。其中，对子边缘线的评分标准与前述对每一条轮廓线的评分标准相同，在此不再重复说明。It can be understood that, for the wafer edge contour line determined in S2, this step cuts the wafer edge contour line into multiple sub-edge lines. For example, the wafer edge contour line is divided into multiple sub-edge lines, and each A sub-edge line is scored, and the optimal multi-segment sub-edge line is selected, that is, the multi-segment sub-edge line with the lowest Score _edge . Wherein, the scoring standard for the sub-edge line is the same as the aforementioned scoring standard for each contour line, and the description is not repeated here.

对于选取的最优的多段子边缘线，从中选取边缘点，比如，从每一段最优的子边缘线上选取若干边缘点，基于选取的所有边缘点，拟合晶圆的圆形形状，可参见图2，得到晶圆的圆心坐标。For the selected optimal sub-edge lines, select edge points from them, for example, select several edge points from each optimal sub-edge line, and fit the circular shape of the wafer based on all the selected edge points. Referring to Figure 2, the coordinates of the center of the wafer are obtained.

实施例二Embodiment 2

一种晶圆缺口定位方法，参见图3，该晶圆缺口定位方法主要包括以下步骤：A wafer gap positioning method, see FIG. 3 , the wafer gap positioning method mainly includes the following steps:

S1’，获取晶圆缺口图像；S1', obtain the wafer gap image;

作为实施例，可以先获取晶圆缺口的光学图像，基于所述晶圆缺口的光学图像，确定晶圆缺口顶端位置。As an embodiment, an optical image of the wafer notch may be obtained first, and the top position of the wafer notch may be determined based on the optical image of the wafer notch.

可以理解的是，为了提高晶圆缺口定位的精度，先对晶圆缺口进行粗定位，然后基于粗定位结果，对晶圆缺口进行细定位，粗定位的目的主要是获取晶圆缺口顶部位置。It can be understood that, in order to improve the accuracy of wafer notch positioning, rough positioning of the wafer notch is performed first, and then based on the coarse positioning result, the wafer notch is finely positioned. The purpose of the rough positioning is mainly to obtain the top position of the wafer notch.

粗定位的步骤主要包括：利用光学探测器获取晶圆缺口的光学图像；对晶圆缺口的光学图像进行去噪预处理；基于去噪预处理后的晶圆缺口的光学图像，对晶圆缺口的光学图像进行二值化处理，比如，可以采用大津法(OSTU)等，从二值化处理后的晶圆缺口的光学图像中提取所有轮廓线。从曲线弯曲程度和近似二次曲线程度两个方面综合评价所有的轮廓线，筛选轮廓线，得到晶圆缺口边缘曲线。The rough positioning steps mainly include: using an optical detector to obtain an optical image of the wafer gap; performing denoising preprocessing on the optical image of the wafer gap; based on the optical image of the wafer gap after denoising For example, the Otsu method (OSTU) can be used to extract all contour lines from the optical image of the wafer notch after binarization. Comprehensively evaluate all the contour lines from the two aspects of curve curvature and approximate quadratic curve degree, screen the contour lines, and obtain the wafer notch edge curve.

可以理解的是，由于光学探测器只能扫描大视野范围内的物体，因此，最终定位出来的晶圆位置不够精确，扫描电子束显微镜SEM能够扫描小视野范围内的晶圆。因此，本发明实施例采用扫描电子束显微镜SEM对晶圆缺口进行扫描，得到晶圆缺口SEM图像。It is understandable that since the optical detector can only scan objects within a large field of view, the position of the wafer finally positioned is not accurate enough. The scanning electron beam microscope (SEM) can scan wafers within a small field of view. Therefore, in the embodiment of the present invention, a scanning electron beam microscope SEM is used to scan the wafer notch to obtain a SEM image of the wafer notch.

对于筛选出的晶圆缺口边缘曲线，选取晶圆缺口边缘曲线中的纵轴最高点为晶圆缺口顶端点位置，获取的晶圆缺口顶部位置见图4。由于晶圆缺口的旋转角度较小，这样选取的点的精度能够满足晶圆缺口粗定位的精度需求，即保证下一步晶圆缺口细定位中晶圆缺口顶端出现在SEM图中。For the screened wafer notch edge curve, the highest point of the vertical axis in the wafer notch edge curve is selected as the position of the top point of the wafer notch, and the obtained top position of the wafer notch is shown in Figure 4. Because the rotation angle of the wafer gap is small, the accuracy of the selected points can meet the precision requirements of the rough positioning of the wafer gap, that is, to ensure that the top of the wafer gap appears in the SEM image in the next step of fine positioning of the wafer gap.

可选的，基于晶圆缺口粗定位过程中获取的晶圆缺口顶端位置，获取晶圆缺口图像。Optionally, the wafer notch image is acquired based on the position of the top of the wafer notch obtained during the rough positioning of the wafer notch.

可以理解的是，基于晶圆缺口粗定位确定的晶圆缺口顶端点位置，利用扫描电子束显微镜扫描晶圆缺口SEM图像，其中，粗定位确定的晶圆缺口顶端出现在晶圆缺口SEM图像中。可选择的，除了以上方法获取晶圆缺口SEM图像，还可以基于其他方法获取晶圆的SEM图像。比如，基于晶圆放置在载台上时晶圆缺口与扫描装置大体确定的位置关系，可以先用扫描装置获取低倍率扫描图像，获取包含晶圆缺口的扫描图像，根据晶圆缺口在低倍率扫描图像中的位置，然后连续调节扫描装置的放大倍率，逐步提高扫描装置的放大倍率，最终获取晶圆缺口的高放大倍率、高分辨率的扫描图像。It is understandable that, based on the position of the top point of the wafer gap determined by the rough positioning of the wafer gap, a scanning electron beam microscope is used to scan the SEM image of the wafer gap, wherein the top of the wafer gap determined by the rough positioning appears in the wafer gap SEM image. . Optionally, in addition to the above method to obtain the SEM image of the wafer notch, the SEM image of the wafer can also be obtained based on other methods. For example, based on the roughly determined positional relationship between the wafer gap and the scanning device when the wafer is placed on the stage, the scanning device can be used to obtain a low-magnification scanned image, and a scanned image containing the wafer gap can be obtained. Scan the position in the image, then continuously adjust the magnification of the scanning device, gradually increase the magnification of the scanning device, and finally obtain a high-magnification, high-resolution scanned image of the wafer gap.

S2’，基于深度学习语义分割模型对所述晶圆缺口图像进行二值化，提取所述晶圆缺口图像中的所有轮廓线。S2', binarize the wafer gap image based on a deep learning semantic segmentation model, and extract all contour lines in the wafer gap image.

可以理解的是，对晶圆缺口SEM图像去噪，将图像大小缩放到深度神经网络(DNN)模型的输入大小，例如512*512，基于深度学习语义分割模型对晶圆缺口SEM图像中进行二值化，从二值化的晶圆缺口SEM图像中获取所有轮廓线。It is understandable that the wafer gap SEM image is denoised, the image size is scaled to the input size of the deep neural network (DNN) model, such as 512*512, and the wafer gap SEM image is divided into two parts based on the deep learning semantic segmentation model. For binarization, all contours were obtained from the binarized wafer notch SEM image.

本发明使用深度学习语义分割模型对晶圆缺口SEM图像进行二值化分割，由于深度语义分割模型提取了不同尺度、不同语义层次的图像特征，分割的边缘更加准确，平滑，且模型的鲁棒性高，泛化能力强，提高了晶圆缺口边缘曲线识别精度。The present invention uses the deep learning semantic segmentation model to perform binarization segmentation on the wafer gap SEM image. Since the deep semantic segmentation model extracts image features of different scales and different semantic levels, the edge of the segmentation is more accurate and smooth, and the model is robust. It has high performance and strong generalization ability, which improves the recognition accuracy of wafer notch edge curve.

S3’，从所有轮廓线中筛选出晶圆缺口边缘曲线。S3', screen out the wafer notch edge curve from all contour lines.

作为实施例，所述从所有轮廓线中筛选出晶圆缺口边缘曲线，包括：提取每一条轮廓线的轮廓特征，所述轮廓特征包括轮廓线的弯曲程度和近似二次曲线程度中任意一个或两个的组合；基于每一条轮廓线的轮廓特征，从所有轮廓线中筛选出晶圆缺口边缘曲线。As an example, the filtering of the wafer notch edge curve from all contour lines includes: extracting contour features of each contour line, the contour features including any one of the degree of curvature of the contour line and the degree of approximate quadratic curve or A combination of the two; based on the contour features of each contour line, the wafer notch edge curve is screened from all contour lines.

可以理解的是，晶圆缺口的轮廓线近似二次曲线，本发明实施例的晶圆缺口的轮廓线的评分方法从曲线的弯曲程度、近似二次曲线程度评价筛选晶圆缺口轮廓线。这种评分筛选轮廓线得到晶圆边缘的方法能够提高模型对晶圆、机台、不同成像质量的图像的泛化能力，从而提高了晶圆缺口的定位精度。It can be understood that the contour of the wafer gap approximates a quadratic curve, and the scoring method of the contour of the wafer gap according to the embodiment of the present invention evaluates and selects the contour of the wafer gap according to the degree of curvature of the curve and the degree of approximate quadratic curve. This method of scoring and screening contour lines to obtain wafer edges can improve the generalization ability of the model to images of wafers, machines, and images with different imaging qualities, thereby improving the positioning accuracy of wafer gaps.

作为实施例，所述提取每一条轮廓线的轮廓特征，包括：根据任一条轮廓线上所有边缘点的横纵坐标的取值范围变化以及轮廓线的二阶梯度，计算所述任一条轮廓线的弯曲程度Score_c；根据所述任一条轮廓线上所有边缘点的横纵坐标，拟合得到二次曲线方程，基于所述二次曲线方程，计算二次曲线拟合的残差平方和SSE_curve作为所述任一条轮廓线的近似二次曲线程度；相应的，基于每一条轮廓线的轮廓特征，从所有轮廓线中筛选出晶圆缺口边缘曲线，包括：所述任一条轮廓线的弯曲程度Score_c以及所述任一条轮廓线的近似二次曲线程度SSE_curve中一个或两个组合的和作为所述任一条轮廓线的评分；将评分最低的轮廓线作为晶圆缺口边缘轮廓线。As an embodiment, the extracting the contour feature of each contour line includes: calculating the any contour line according to the value range change of the abscissa and vertical coordinates of all edge points on any contour line and the second-order gradient of the contour line The degree of curvature Score _c ; according to the abscissa and ordinate of all edge points on the described any contour line, fitting obtains a quadratic curve equation, based on the quadratic curve equation, calculate the residual sum of squares SSE of quadratic curve fitting The _curve is used as the approximate quadratic curve degree of any one of the contour lines; correspondingly, based on the contour features of each contour line, the edge curve of the wafer gap is screened from all the contour lines, including: the bending of any one of the contour lines. The degree Score _c and the sum of one or two combinations of the approximate quadratic degree SSE _curve of any contour line are used as the score of any one contour line; the contour line with the lowest score is regarded as the wafer notch edge contour line.

具体的，对于从晶圆缺口SEM图像中提取的每一条轮廓线，计算其弯曲程度和近似二次曲线程度，其中，轮廓线的弯曲程度Score_c可通过公式(10)来衡量。对于近似二次曲线程度的计算，由于晶圆缺口曲线近似于二次曲线，因此，基于轮廓线上的边缘点进行二次曲线拟合：Specifically, for each contour line extracted from the wafer notch SEM image, the degree of curvature and the degree of approximate quadratic curve are calculated, wherein the degree of curvature of the contour line, Score _c , can be measured by formula (10). For the calculation of the degree of approximate quadratic curve, since the wafer notch curve approximates the quadratic curve, the quadratic curve fitting is performed based on the edge points on the contour:

y_curve＝w₀+w₁x+w₂x² (14)；y _curve = w ₀ +w ₁ x+w ₂ x ² (14);

上式为晶圆缺口曲线的拟合公式，用最小二乘法拟合得到

The above formula is the fitting formula of the wafer notch curve, which is obtained by the least squares method

其中，SSE_curve为最小二乘拟合得到的一元二次方程的残差平方和，作为近似二次曲线程度。Among them, SSE _curve is the residual sum of squares of the quadratic equation obtained by the least squares fitting, as the approximate quadratic curve degree.

Score_notch＝λ_cScore_c+λ_curveSSE_curve (16)；Score _notch =λ _c Score _c +λ _curve SSE _curve (16);

其中，λ_c、λ_curve分别为Score_c和SSE_curve的权重，Score_notch为晶圆缺口轮廓线的评分公式。本发明实施例中，λ_c、λ_curve均取1。Among them, λ _c and λ _curve are the weights of Score _c and SSE _curve respectively, and Score _notch is the scoring formula of the wafer notch contour. In the embodiment of the present invention, λ _c and λ _curve both take 1.

对于每一条晶圆缺口轮廓线，根据公式(16)对其进行评分，得到对应的分值，将分值最低的轮廓线作为晶圆缺口边缘曲线。For each wafer notch contour line, score it according to formula (16) to obtain the corresponding score, and take the contour line with the lowest score as the wafer notch edge curve.

在另一实施例中，可以选择基于轮廓线的弯曲程度或者近似二次曲线程度之一作为评分依据，从所有轮廓线中筛选出晶圆缺口边缘曲线。以采用基于轮廓线近似二次曲线程度作为评分依据，采用上述相同方法获取任一轮廓线近似二次曲线程度SSE_curve，此时评分为Score_notch＝SSE_curve，将分值最低的轮廓线作为晶圆缺口边缘曲线。In another embodiment, one of the degree of curvature of the contour line or the degree of approximate quadratic curve may be selected as the scoring basis, and the wafer notch edge curve is screened from all the contour lines. Taking the degree of approximate quadratic curve based on the contour line as the scoring basis, the same method as above is used to obtain the approximate quadratic curve degree SSE _curve of any contour line. At this time, the score is Score _notch = SSE _curve , and the contour line with the lowest score is used as the Round notch edge curve.

以采用基于轮廓线弯曲程度作为评分依据，采用上述相同方法获取任一轮廓线的弯曲程序Score_c，此时评分为Score_notch＝Score_c，将评分最低的轮廓线作为晶圆缺口边缘曲线。Taking the degree of curvature based on the contour as the scoring basis, the same method as above is used to obtain the bending program Score _c of any contour line, at this time the score is Score _notch = Score _c , and the contour line with the lowest score is used as the wafer notch edge curve.

S4’，从所述晶圆缺口边缘曲线中确定晶圆缺口的顶部边缘曲线。S4', determining the top edge curve of the wafer gap from the wafer gap edge curve.

通过上述步骤S3’获取晶圆缺口边缘曲线，本步骤获取所述晶圆缺口边缘曲线的所有边缘点的中心点，以所述中心点为基准，截取所述晶圆缺口边缘曲线，将所述晶圆缺口边缘曲线两侧边缘部分剔除，得到所述晶圆缺口的顶部边缘曲线。示例性的，以所有边缘点的中心点为基准，左右两侧对称截取晶圆缺口边缘曲线，将两侧对称的边缘部分剔除，仅保留中心点为基准的曲线部分即为缺口的顶部边缘曲线。如图5所示，图中虚线框为获取的晶圆缺口的顶部边缘曲线。Obtaining the wafer notch edge curve through the above step S3', this step obtains the center points of all edge points of the wafer notch edge curve, taking the center point as a benchmark, intercepting the wafer notch edge curve, and using the The edge portions on both sides of the edge curve of the wafer gap are removed to obtain the top edge curve of the wafer gap. Exemplarily, based on the center point of all edge points, the edge curve of the wafer notch is symmetrically intercepted on the left and right sides, the edge parts symmetrical on both sides are removed, and only the curve part with the center point as the reference is retained as the top edge curve of the notch. . As shown in FIG. 5 , the dotted box in the figure is the obtained top edge curve of the wafer gap.

S5’，基于晶圆缺口的顶部边缘曲线，拟合得到晶圆缺口圆心坐标。S5', based on the top edge curve of the wafer gap, fit the coordinates of the center of the wafer gap.

作为实施例，所述基于晶圆缺口的顶部边缘曲线，拟合得到晶圆缺口圆心坐标，包括：获取顶部边缘曲线上的所有顶部边缘点，构建多个顶部边缘点集合，每个顶部边缘点集合包含的顶部边缘点不重合，对每个顶部边缘点集合中的边缘点，拟合得到对应晶圆缺口圆心坐标；将所有晶圆缺口圆心坐标求取平均值，得到最终的晶圆缺口圆心坐标。如图5所示，基于虚线框内的晶圆顶部缺口的顶部边缘曲线，拟合得到对应的圆，该圆的圆心坐标即为晶圆缺口的圆心坐标，示例性的，图5仅示出了一个顶部边缘点集合的拟合结果。As an example, the fitting to obtain the coordinates of the center of the wafer gap based on the top edge curve of the wafer gap includes: acquiring all top edge points on the top edge curve, constructing a plurality of top edge point sets, each top edge point The top edge points contained in the set do not overlap. For each edge point in the top edge point set, the corresponding wafer gap center coordinates are obtained by fitting; the average value of all the wafer gap center coordinates is obtained to obtain the final wafer gap center coordinate. As shown in FIG. 5 , based on the top edge curve of the wafer top notch in the dotted frame, a corresponding circle is obtained by fitting, and the center coordinate of the circle is the center coordinate of the wafer notch. For example, FIG. 5 only shows The fitting result of a set of top edge points.

可以理解的是，由于晶圆缺口处可能存在缺陷，图像二值化后筛选得到的晶圆缺口轮廓线可能有凹凸，不够平滑，导致拟合的圆不够准确。本发明通过去除缺口边缘曲线的边缘部分，通过基于晶圆缺口的顶部边缘曲线，选取多组顶部边缘点集合，分别拟合，得到多个晶圆缺口圆心坐标，最后对圆心点集中所有的圆心坐标取平均得到晶圆缺口圆心坐标。这种方法能够增加方法寻找晶圆缺口圆心的准确性和鲁棒性。It is understandable that, due to possible defects in the wafer gap, the wafer gap contour line screened after image binarization may be uneven and not smooth enough, resulting in an inaccurate fitted circle. In the present invention, by removing the edge part of the edge curve of the notch, and by selecting a plurality of sets of top edge points based on the top edge curve of the wafer notch, and fitting them respectively, the coordinates of the center of the multiple wafer notch are obtained, and finally all the center points are collected for the center point. The coordinates are averaged to obtain the coordinates of the center of the wafer gap. This method can increase the accuracy and robustness of the method for finding the center of the wafer notch.

实施例三Embodiment 3

一种晶圆定位校准方法，需要获取晶圆圆心坐标和晶圆缺口的圆心坐标。其中，晶圆圆心坐标可由实施例一的方法获取，晶圆缺口的圆心坐标可由实施例二的方法获取。A wafer positioning calibration method needs to obtain the coordinates of the center of the wafer and the coordinates of the center of the wafer gap. The coordinates of the center of the wafer can be obtained by the method of the first embodiment, and the coordinates of the center of the wafer notch can be obtained by the method of the second embodiment.

基于晶圆圆心坐标和晶圆缺口的圆心坐标，计算得到晶圆旋转角，从而完成晶圆定位校准。Based on the coordinates of the center of the wafer and the coordinates of the center of the wafer gap, the rotation angle of the wafer is calculated to complete the wafer positioning calibration.

由晶圆圆心坐标和晶圆缺口的圆心坐标，可以得到两个圆心坐标连线与晶圆载台坐标系y轴方向上的夹角，即为晶圆的旋转角，从而可以调整晶圆的位置和角度，完成晶圆的定位校准。From the coordinates of the wafer center and the center coordinates of the wafer gap, the angle between the two center coordinates and the y-axis direction of the wafer stage coordinate system can be obtained, which is the rotation angle of the wafer, so that the wafer can be adjusted. position and angle to complete the positioning and calibration of the wafer.

实施例四Embodiment 4

一种晶圆圆心定位系统，参见图6，该晶圆圆心定位系统包括第一获取模块601、第一提取模块602、第一筛选模块603和第一拟合模块604。A wafer center positioning system, see FIG. 6 , the wafer center positioning system includes a first acquisition module 601 , a first extraction module 602 , a first screening module 603 and a first fitting module 604 .

其中，第一获取模块601，用于获取晶圆圆周图像；第一提取模块602，用于基于深度学习语义分割模型对所述晶圆圆周图像进行二值化处理，提取所述晶圆圆周图像中的所有轮廓线；第一筛选模块603，用于从所有轮廓线中筛选出晶圆边缘轮廓线；第一拟合模块604，用于基于所述晶圆边缘轮廓线，拟合得到晶圆圆心坐标。Among them, the first acquisition module 601 is used for acquiring the wafer circumference image; the first extraction module 602 is used for binarizing the wafer circumference image based on the deep learning semantic segmentation model, and extracting the wafer circumference image The first screening module 603 is used to screen out the wafer edge contour lines from all the contour lines; the first fitting module 604 is used to obtain the wafer by fitting based on the wafer edge contour lines The coordinates of the center of the circle.

可以理解的是，本发明提供的一种晶圆圆心定位系统与前述各实施例提供的晶圆圆心定位方法相对应，晶圆圆心定位系统的相关技术特征可参考晶圆圆心定位方法的相关技术特征，在此不再赘述。It can be understood that the wafer center positioning system provided by the present invention corresponds to the wafer center positioning methods provided in the foregoing embodiments, and the related technical features of the wafer center positioning system may refer to the related technologies of the wafer center positioning method. features, which will not be repeated here.

实施例五Embodiment 5

一种晶圆缺口定位系统，参见图7，该晶圆缺口定位系统包括第二获取模块701、第二提取模块702、第二筛选模块703、确定模块704和第二拟合模块705。A wafer gap positioning system, see FIG. 7 , the wafer gap positioning system includes a second acquisition module 701 , a second extraction module 702 , a second screening module 703 , a determination module 704 and a second fitting module 705 .

其中，第二获取模块701，用于获取晶圆缺口图像；第二提取模块702，用于基于深度学习语义分割模型对所述晶圆缺口图像进行二值化，提取所述晶圆缺口图像中的所有轮廓线；第二筛选模块703，用于从所有轮廓线中筛选出晶圆缺口边缘曲线；确定模块704，用于从所述晶圆缺口边缘曲线确定晶圆缺口的顶部边缘曲线；第二拟合模块705，用于基于所述晶圆缺口的顶部边缘曲线，拟合得到晶圆缺口圆心坐标。Wherein, the second acquisition module 701 is used to acquire the wafer gap image; the second extraction module 702 is used to binarize the wafer gap image based on a deep learning semantic segmentation model, and extract the wafer gap image from the wafer gap image. The second screening module 703 is used to screen out the edge curve of the wafer gap from all the contour lines; the determining module 704 is used to determine the top edge curve of the wafer gap from the edge curve of the wafer gap; the first The second fitting module 705 is configured to obtain the coordinates of the center of the wafer notch by fitting based on the top edge curve of the wafer notch.

可以理解的是，本发明提供的一种晶圆缺口定位系统与前述各实施例提供的晶圆缺口定位方法相对应，晶圆缺口定位系统的相关技术特征可参考晶圆缺口定位方法的相关技术特征，在此不再赘述。It can be understood that a wafer gap positioning system provided by the present invention corresponds to the wafer gap positioning methods provided in the foregoing embodiments, and the related technical features of the wafer gap positioning system may refer to the related technologies of the wafer gap positioning method. features, which will not be repeated here.

实施例六Embodiment 6

请参阅图8，图8为本发明实施例提供的电子设备的实施例示意图。如图8所示，本发明实施例提了一种电子设备800，包括存储器810、处理器820及存储在存储810上并可在处理器820上运行的计算机程序811，处理器820执行计算机程序811时实现实施例一的晶圆圆心定位方法、实施例二的晶圆缺口定位方法或实施例三的晶圆定位校准方法。Please refer to FIG. 8 , which is a schematic diagram of an embodiment of an electronic device provided by an embodiment of the present invention. As shown in FIG. 8 , an embodiment of the present invention provides an electronic device 800, including a memory 810, a processor 820, and a computer program 811 stored in the storage 810 and running on the processor 820, and the processor 820 executes the computer program At 811, the wafer center positioning method of the first embodiment, the wafer gap positioning method of the second embodiment, or the wafer positioning calibration method of the third embodiment are implemented.

实施例七Embodiment 7

请参阅图9，图9为本发明提供的一种计算机可读存储介质的实施例示意图。如图9所示，本实施例提供了一种计算机可读存储介质900，其上存储有计算机程序911，该计算机程序911被处理器执行时实现实施例一的晶圆圆心定位方法或实施例二的晶圆缺口定位方法、实施例三的的晶圆定位校准方法。Please refer to FIG. 9, which is a schematic diagram of an embodiment of a computer-readable storage medium provided by the present invention. As shown in FIG. 9 , this embodiment provides a computer-readable storage medium 900 on which a computer program 911 is stored. When the computer program 911 is executed by a processor, the method or the embodiment for locating the wafer center of the first embodiment is implemented. The second wafer notch positioning method, and the third embodiment of the wafer positioning calibration method.

本发明实施例提供的一种晶圆圆心定位、晶圆缺口定位及晶圆定位校准方法，获取晶圆边缘的SEM图像，使用深度学习语义分割模型对图像做二值化处理，通过评分筛选轮廓线得到晶圆边缘，用最小二乘法拟合晶圆圆心；使用光学探测器对晶圆缺口粗定位，利用粗定位的结果获取晶圆缺口SEM图，使用深度学习语义分割模型对图像做二值化处理，通过评分筛选轮廓线得到晶圆缺口边缘曲线，拟合晶圆缺口圆心完成晶圆缺口细定位；以及基于获取的晶圆圆心和晶圆缺口圆心，对晶圆定位进行校准。本发明公开了一种基于扫描电子束显微镜(SEM)和深度学习(Deep learning)的晶圆定位方法，可以精确的测量晶圆圆心和缺口；提高了晶圆定位的精度，将定位精度提升两个数量级，从100μm提升到5μm，且无需人为干预或其它先验信息，能自动测量晶圆圆心和缺口。The embodiment of the present invention provides a wafer center positioning, wafer gap positioning and wafer positioning calibration method, which acquires a SEM image of the wafer edge, uses a deep learning semantic segmentation model to binarize the image, and selects the contour by scoring. The edge of the wafer is obtained from the line, and the center of the wafer is fitted by the least square method; the optical detector is used to roughly locate the wafer gap, and the SEM image of the wafer gap is obtained by using the result of the rough positioning, and the image is binarized using the deep learning semantic segmentation model. The wafer notch edge curve is obtained by scoring and screening contour lines, and the wafer notch center is fitted to complete the wafer notch fine positioning; and the wafer positioning is calibrated based on the obtained wafer center and wafer notch center. The invention discloses a wafer positioning method based on scanning electron beam microscope (SEM) and deep learning (Deep learning), which can accurately measure the wafer center and gap; improve the wafer positioning accuracy, and improve the positioning accuracy by two An order of magnitude, from 100μm to 5μm, and can automatically measure the wafer center and gap without human intervention or other prior information.

需要说明的是，在上述实施例中，对各个实施例的描述都各有侧重，某个实施例中没有详细描述的部分，可以参见其它实施例的相关描述。It should be noted that, in the foregoing embodiments, the description of each embodiment has its own emphasis, and for parts that are not described in detail in a certain embodiment, reference may be made to the relevant descriptions of other embodiments.

本领域内的技术人员应明白，本发明的实施例可提供为方法、系统、或计算机程序产品。因此，本发明可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本发明可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。As will be appreciated by one skilled in the art, embodiments of the present invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

本发明是参照根据本发明实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式计算机或者其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。The present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each flow and/or block in the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to the processor of a general purpose computer, special purpose computer, embedded computer or other programmable data processing device to produce a machine such that the instructions executed by the processor of the computer or other programmable data processing device produce Means implementing the functions specified in one or more of the flowcharts and/or one or more blocks of the block diagrams.

这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory result in an article of manufacture comprising instruction means, the instructions The apparatus implements the functions specified in the flow or flow of the flowcharts and/or the block or blocks of the block diagrams.

这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。These computer program instructions can also be loaded on a computer or other programmable data processing device to cause a series of operational steps to be performed on the computer or other programmable device to produce a computer-implemented process such that The instructions provide steps for implementing the functions specified in the flow or blocks of the flowcharts and/or the block or blocks of the block diagrams.

尽管已描述了本发明的优选实施例，但本领域内的技术人员一旦得知了基本创造概念，则可对这些实施例作出另外的变更和修改。所以，所附权利要求意欲解释为包括优选实施例以及落入本发明范围的所有变更和修改。Although preferred embodiments of the present invention have been described, additional changes and modifications to these embodiments may occur to those skilled in the art once the basic inventive concepts are known. Therefore, the appended claims are intended to be construed to include the preferred embodiment and all changes and modifications that fall within the scope of the present invention.

显然，本领域的技术人员可以对本发明进行各种改动和变型而不脱离本发明的精神和范围。这样，倘若本发明的这些修改和变型属于本发明权利要求及其等同技术的范围之内，则本发明也意图包括这些改动和变型在内。It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit and scope of the invention. Thus, provided that these modifications and variations of the present invention fall within the scope of the claims of the present invention and their equivalents, the present invention is also intended to include such modifications and variations.